Skip to main content
NCBI home page
BMC Pediatrics logoLink to BMC Pediatrics
. 2026 Jan 26;26:126. doi: 10.1186/s12887-026-06512-3

Predicting neuroblastoma prognosis using machine learning analysis of metabolism-related gene network perturbation

Xin Liu 1,#, Xin Hu 2,#, Qinzhen Cai 3,#, Chunhui Yuan 3,✉, Jun Wang 3,✉
PMCID: PMC12915021  PMID: 41588361

Abstract

Background

Neuroblastomas (NB) are highly heterogeneous pediatric extracranial solid tumors in children with variable epigenetic, biological, and clinical characteristics. However, predictive models that can accurately classify patient risks and predict prognoses are currently limited. We analyzed a metabolism-related genes network perturbation using machine learning algorithms to construct a model to assess the risk and prognosis of patients with NB.

Methods

Metabolism-related gene expression data for patients with NB were obtained from the Gene Expression Omnibus (GEO) (GSE49710, N = 498), ArrayExpress (E-MTAB-8248, N = 228), and TARGET (TARGET-NBL, N = 150) databases. An individual-specific gene interaction perturbation network was constructed using the Reactome Pathway Database. Unsupervised clusters and principal components analysis were analyzed using the R package “Consensus Cluster Plus”. Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analyses were conducted using the R package “Cluster Profiler”. Gene set variation analysis (GSVA) and single-sample gene set enrichment analysis (ssGSEA) were performed to evaluate immune cell infiltration in patients with NB using the HALLMARK database. Ten machine learning methods (Lasso, Enet, plsRcox, CoxBoost, StepCox, gradient boosting machine [GBM], Ridge, random survival forests [RSF], survival-support vector machine [SVM], and super principal component [PC]) were used with 110 machine learning algorithms to screen for a metabolism-related signature to predict NB prognosis in GSE49710 and other cohorts (E-MTAB-8248 and GSE76427). Finally, Immunohistochemical staining (IHC) was performed to validate the UCK2 expression levels in human tissues.

Results

Gene network perturbation analysis of 948 metabolism-related genes revealed distinct discriminability, and patients with NB were classified into three differentiated subtypes. Kaplan-Meier survival analysis revealed that the prognoses were best for patients with subtype C3, followed by those with subtypes C1 and C2. Correlation analysis of clinical information indicated that subtypes C2 and C3 were associated with higher and lower percentages of high-degree malignancies, respectively, whereas C3 subtype showed a lower percentage of high-degree malignancies. The KRAS and myogenesis pathways were upregulated, and the levels of MYC targets were downregulated in patients with subtype C3; those with C2 subtype exhibited opposite trends. Patients with C3- and C2-subtypes exhibited immune-activated and immune-suppressed phenotypes, respectively. The combination of the StepCox [forward] and RSF algorithms provided the most accurate prognostic predictions for patients with NB. The importance score was highest for UCK2 among all subtypes. IHC staining further confirmed that UCK2 expression was substantially higher in the tissues of patients with NB than in those of controls.

Conclusions

The machine learning-based prognostic prediction model that analyzes metabolism-related gene interaction perturbation networks supports the development of personalized management strategies for patients with NB.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12887-026-06512-3.

Keywords: Neuroblastoma, Metabolism, Gene network perturbation, Unsupervised cluster, Machine learning, Prognosis

Introduction

Neuroblastomas (NB) are common extracranial solid tumors in children, accounting for approximately 7% of all pediatric cancer cases and contributing to approximately 15% of pediatric cancer-related mortality worldwide [1]. NB is a highly heterogeneous disease with variable epigenetic, biological, and clinical characteristics, such as widespread metastasis, high aggressiveness, resistance to therapy, long-term treatment toxicities, and an elevated risk of recurrence [2]. Advances in NB pathogenesis and the identification of key drivers of high-risk have gradually increased the efficacy of multimodal therapies, leading to a 5-year survival rate of approximately 50% for high-risk patients [2, 3]. However, these improvements in therapeutic efficacy are not sufficient to meet the needs of clinicians and patients [4]. Therefore, further studies to stratify patient risk, optimize appropriate therapeutic regimens, and provide early prediction and intervention for poor prognosis are required to improve the outcomes of patients with NB.

Several molecular biomarkers of NB have been identified and have been used in clinical trials to preliminarily assess its occurrence, progression, treatment efficacy, prognosis, and recurrence. Urinary catecholamine metabolites such as homovanillic acid or vanillylmandelic acid are relatively simple and widely used biomarkers for diagnosing and monitoring NB recurrence [5, 6]. Oncogene MYCN amplification, TERT alterations, and mutations in the RAS and p53 pathways are often associated with poor survival outcomes [7, 8] and may influence the effectiveness of immunotherapies [9]. ALK mutations are potential targets for precision medicine approaches [10, 11]. Despite these advancements, no single biomarker can be used to independently classify or predict NB outcomes. Consequently, combining age, disease stage, and MYCN amplification status remains the cornerstone for stratifying the risk and evaluating the prognosis of NB [2], which significantly limits the clinical diagnostic and therapeutic strategies. Thus, identifying novel biomarkers based on a detailed understanding of NB cell progression is clinically valuable for enhancing diagnosis and treatment.

Tumorigenesis, such as that which occurs in NB, is a highly energy-consuming process that requires efficient energy metabolism to support the proliferation of malignant cells [12]. Proteomics analyses of NB have confirmed that numerous significant metabolic enzymes are involved in tumor-dependent modifications of metabolic networks within the tumor microenvironment (TME). These enzymes contribute to key energy metabolic pathways, such as lipolysis and fatty acid β-oxidation, glycolysis, the tricarboxylic acid cycle, and oxidative phosphorylation, mediating NB tumorigenesis, angiogenesis, drug resistance, and tumor-induced immune suppression [12–14]. However, the synergistic roles of these pathways are complex and remain largely unknown. Therefore, understanding how these metabolism-related biomarkers change in patients with NB will provide important information for a more accurately prediction of NB progression, classification of patient risk, development of precision therapy, and monitoring of prognosis.

We developed an unsupervised clustering method using gene network perturbation to analyze clusters of NB metabolism-related genes from multiple databases to understand NB progression and identify metabolism-related biomarkers for risk stratification and prognostic monitoring. In addition, we constructed a prognostic prediction model using 110 machine learning (ML) algorithms based on these potential biomarkers. This study provides a comprehensive overview of the prognosis, biological functions, and characteristics of the tumor immune microenvironment (TIME) between clusters. Therefore, our study provides a high-precision classification system for patients with NB and enhances our understanding of the heterogeneity of gene interaction networks in NB.

Materials and methods

Data acquisition and collection

NB genes expression data were sourced from the Gene Expression Omnibus (GEO) database (https://www.ncbi.nlm.nih.gov/geo/, GSE49710, N = 498), the ArrayExpress database (https://www.ebi.ac.uk/arrayexpress, E-MTAB-8248, N = 228), and the TARGET database (https://targetbase.com/, TARGET-NBL, N = 150), including samples with both expression and survival data. 948 metabolism-related genes and 41 pathways were obtained from the MsigDb database (https://www.gsea-msigdb.org/gsea/msigdb) (Supplementary Table 1), with transcripts per million (TPM) values log2(TPM + 1) transformed and Z-scaled. GEO database (GSE73517, N = 105) was used to explore links between classifications and clinical information.

Gene perturbation network model construction

The Reactome Pathway Database (https://reactome.org/) covers over 50% of human proteins, providing protein interaction networks. Reactions are the core of Reactome’s data model, with entities like proteins and small molecules forming interactions networks categorized into pathways. To model gene perturbation networks, gene associations are mapped onto theses interaction networks for pathway-based analysis [15]. The networks are organized into pathway names and associated coding genes. These individual networks are consolidated into a single comprehensive network with 167,849 edges. Our methodology involves three main steps, as shown in Fig. 1. First, a gene expression rank matrix is created, ranking each gene expression value in each sample (Gn). Next, a δ-rank matrix is calculated, representing the rank differences between two connected genes, modeled as edges in a gene perturbation network (En). Finally, a baseline δ-rank vector is established from normal samples. The side perturbation matrix is then generated by subtracting each gene’s rank in the δ-rank matrix from the baseline δ-rank vector. The δ-rank matrix stores the rank difference of each interacting gene pair; larger differences indicate stronger edge perturbation. This matrix effectively quantifies sample-specific gene interaction perturbations within a consistent network, with each column reflecting the perturbation for an individual sample.

Fig. 1.

Fig. 1

Open in a new tab

Schematic diagram of NB metabolism-related genes network perturbation analysis

Unsupervised cluster analysis

Consensus cluster is a robust unsupervised learning tool that uses consensus clustering and principal component analysis on high-dimensional single nucleotide polymorphism (SNP) and gene expression data to achieve precise and stable clustering, revealing hidden biological interpretation [16]. Unsupervised cluster analysis is performed using “Consensus Cluster Plus” R package, with a resampling parameter set to 1,000. The stability of the classification is ensured through the application of the K-means algorithm, executed with 1,000 iterations and 80% resampling. Subsequently, Principal Component Analysis (PCA) maps are generated to further elucidate the distinctions between clusters.

Gene enrichment analysis

Genomic files from the Hallmark database (https://www.kegg.ip/) were analyzed for variants using the “GSEA Base” package (version 1.44.0) in R. Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analyses were conducted with “Cluster Profiler” package, applying a significance threshold of P < 0.05. Results were displayed as a heatmap, and figures were reproduced following KEGG’s copyright policy with appropriate citation [17].

Quantitative analysis of tumor-infiltrating immune microenvironment

The analysis of enrichment scores effectively mitigates the limitation of focusing solely on the most and least expressed genes with significant inter-group differences, thereby preventing the neglect of comprehensive genomic information and enhancing the stability and accuracy of enrichment outcomes [18].The TPM format of NB mRNA expression data, the “Filter Common Genes” and “Estimate Score” functions from the R package ESTIMATE, were utilized to calculate the matrix fraction and immune fraction across various subtypes, facilitating the characterization of differences in TIME. The immune checkpoint target-related genes and the definition of cell types within the tumor microenvironment is grounded in previous studies [19, 20]. The expression levels of 24 immune cell infiltration in various subtypes were quantified using the “GSVA” R software package. Specifically, single-sample Gene Set Enrichment Analysis (ssGSEA) was applied to assess immune cell infiltration across the subtypes.

Multiple ML algorithms to construct an efficient NB prognosis prediction model

Univariate Cox regression analysis was initially employed to identify metabolism-related genes affecting NB prognosis. Genes with significant differences (P < 0.05) were subjected to further analysis using a combination of 110 algorithmic approaches across 10 ML methods, including Lasso, Enet, plsRcox, CoxBoost, StepCox, GBM, Ridge, RSF, Survivor-SVM, and SuperPC. These methods were applied to screen for signature genes in the GSE49710 dataset and additional cohorts (E-MTAB-8248 and GSE76427). Each algorithm calculated the mean concordance index (C-index), and threshold values derived from the survminer R package were used to categorize NB patients into high- and low-risk groups. Kaplan-Meier (K-M) survival analysis and Receiver Operating Characteristic (ROC) curve were employed to assess the OS efficiency and survival prediction accuracy at 1-, 3-, and 5-year within the GSE49710 cohort. The Area Under the Curve (AUC) values were subsequently calculated for both training and validation datasets. Finally, the models were ranked based on their average AUC values to identify the optimal models. The accuracy of the models was further assessed using ROC curves. Penalized Cox models including Lasso, Enet, and Ridge, employed nested 10-fold cross-validation to select the penalty parameter λ at 1 standard error.

NB patients tissue collection and immunohistochemical staining

10 NB patient’s tumor tissues and 5 Ganglioneuroma patient’s tissues were collected from September 2021 to June 2022 in Department of General Surgery, Wuhan Children’s Hospital. The inclusion criteria: confirmed as NB or ganglioneuroma children by pathological diagnosis. Exclusion criteria included: (1) Taking drugs that may affect the immune function of the child; (2) Other autoimmune, infectious, neoplastic and other related diseases; (3) The patients were treated with chemoradiotherapy before admission; (4) Incomplete information of children. The study protocol received approval from the Ethics Committee of Wuhan Children’s Hospital (2021R134-E01), and all procedures were conducted in compliance with the ethical standards outlined in the Declaration of Helsinki. Written informed consent was obtained from all participants or their guardians before study enrollment. Tissues immunohistochemical staining was conducted following the standard laboratory procedures. Initially, the tissue samples were fixed using paraformaldehyde, and a series of graded ethanol concentrations, followed by embedding in liquid paraffin to produce sections with a thickness of 5 mm. The slides were then treated with a nonspecific sodium citrate antigen repair solution (pH 6.0), 3% bovine serum albumin before undergoing sequential incubation with an anti-UCK2 antibody (10511-1-AP, Proteintech) and HRP-conjugated secondary antibodies (SA00001-2, Proteintech). Subsequently, 3,3’ -diaminobenzidine and hematoxylin were applied. The stained sections were examined for positive cells identification using a BX51 microscope (Olympus).

Statistical analysis

Statistical analysis and visualization in the bioinformatics section were conducted by R software (Version 4.3.2). Categorical variables were analyzed with Fisher’s exact test, while univariate Cox regression assessed the link between metabolism-related genes and NB patient prognosis. The Wilcoxon test compared two groups, and the Kruskal-Wallis test was used for three or more groups. Kaplan-Meier analysis generated survival curves, with differences analyzed by the log-rank method. Significant was set at P < 0.05. Micro-array batch effects were removed using ComBat (sva v3.46.0); and RNA-seq data were adjusted with limma-voom, including “batch” as a covariate. Probes/genes with > 5% missing values were imputed with k-nearest neighbor (k = 10, impute v1.72.3). A full grid search was performed in a 5-fold inner CV for 110 algorithm combinations, and an outer 5-fold CV obtained an unbiased C-index. Final performance is reported as the mean C-index with a 95% CI from 1,000 bootstrap samples.

Results

Metabolism-related genes network perturbation analysis revealed different subtypes in NB

Through the 948 metabolism-related genes, the perturbation network matrix of the metabolism-related genes was calculated based on the interaction relationships of the perturbation network. The Final Matrix of metabolism-related gene interactions in the NB was obtained (Fig. 1). Unsupervised clustering was used to analyze the perturbed network subtypes of potentially interacting genes in NB. The results showed that metabolism-related genes could classify patients with NB into three subtypes (Fig. 2A). All samples were projected onto two-dimensional spatial coordinates using the Unified Manifold Approximation and Projection (UMAP), which showed that the network perturbation analysis had good discriminability (Fig. 2B). To further explore the biological functions of the NB molecular subtypes, we performed GSVA enrichment analysis to analyze 41 pathways in the Hallmark database among the three subtypes, purine metabolism, pyrimidine metabolism, cysteine and methionine metabolism, and glyoxylate and dicarboxylate metabolic pathways, which were significantly activated in the C2 subtype, while the inositol phosphate metabolism, glycerolipid metabolism, and glycerophospholipid metabolism pathways were significantly activated in the C3 subtype (Fig. 2C). These results reveal that the NB patients have significant molecular convergence, and metabolism-related genes network perturbation may have a rich value in NB discrimination.

Fig. 2.

Fig. 2

Open in a new tab

Metabolism-related genes network perturbation analysis effectively classified NB patients into three distinct subtypes (C1, C2, and C3). A NB patients were divided into three subtypes by metabolism-related genes unsupervised cluster analysis. B PCA analysis demonstrated the independence among the C1, C2, and C3 subtypes. C GSVA enrichment analysis revealed the activation of 41 metabolic pathways (KEGG pathway), 948 metabolic genes across the three subtypes. Different colors represented different activation pathways in heat map, with yellow represented activation of the pathways and blue represented inhibition of the pathways. The red font highlights the four most significantly activated metabolic pathways in the C2 subtype, while the pink font highlights the three most significantly activated metabolic pathways in the C3 subtype

Different subtypes present different NB prognostics

To further analyze the prognosis of patients with NB among the different subtypes, we conducted a Kaplan-Meier survival analysis on the GSE49710 cohort, which showed that patients in the C2 subtype had a poor survival probability, whereas those in the C3 subtype had a relatively good prognosis, achieving more clinical benefit. The prognosis of patients with C1 subtypes was intermediate (Fig. 3A). Multivariate COX regression analysis and random forest plots showed that the C3 subtype had the best prognosis and while C2 subtype had the worst prognosis (HR = 27.853, 95%CI = 11.206–69.41, P < 0.001) (Fig. 3B). To further verify the reliability of the classification, two external NB cohorts (AE (N = 223) and TARGET (N = 150)) were used and the results were consistent with those of the GSE49710 cohort. C3 showed the best prognosis, followed by subtypes C1 and C2 (AE, P < 0.001, Fig. 3C; TARGET, P = 0.003, Fig. 3D). Therefore, these preliminary findings further indicate that the metabolism-related genes can effectively predict the prognosis of NB patients.

Fig. 3.

Fig. 3

Open in a new tab

The prognostic differences among the three subtypes. A Kaplan-Meier survival curve illustrated the survival status of C1, C2, and C3 subtypes in GSE49710 cohort. B Random Forest plots displayed multivariate COX regression analysis of Hazard ratio among three subtypes in GSE49710 cohort. C Kaplan-Meier survival curve showed the survival status of C1, C2, and C3 subtypes in AE cohort. D Kaplan-Meier survival curve showed the survival status of C1, C2, and C3 subtypes in TARGET cohort. **: P < 0.01; ***= P < 0.001

Clinical information’s correlation analysis among the three subtypes

To further validate the accuracy of the metabolism-related genes for NB classification, we analyzed the association between the three subtypes and NB clinical indicators in the GSE73517 cohort, and the results showed statistical differences. Stages analysis showed that C2 subtypes had a high percentage of high degrees of malignancy (Stages 3 and 4), whereas C3 subtypes had a low percentage of high degrees of malignancy (Stages 1 and 4 S) (Fig. 4A, P = 0.0004998). Risk score analysis showed the C2 subtypes had a high percentage of high-risk, whereas > 50% of C3 subtypes were from the low-risk group (Fig. 4B, P = 2.982e-10). MYCN amplification analysis revealed that the C2 subtype had > 60% MYCN amplification, whereas the C1 and C3 subtypes showed lower percent MYCN amplification, particularly the C3 subtype (Fig. 4C, P = 6.811e-11). Somatic genetic analysis showed that the C3 subtype had a lower 17q gain (Fig. 4D, P = 2.608e-5), a higher percentage of 11q whole loss (Fig. 4E, P = 0.0007136), and was 1p normal (Fig. 4F, P = 4.027e-6), whereas the C2 subtypes had the opposite results. In summary, the clinical information for the three subtypes was generally consistent with their prognoses.

Fig. 4.

Fig. 4

Open in a new tab

Correlation analysis of clinical information among the three subtypes. A Percentage distribution of different stages among the three subtypes in GSE73517 cohort. B Percentage distribution of risks cores analysi among the three subtypes in the GSE73517 cohort. C Percentage distribution of MYCN amplification among the three subtypes in the GSE73517 cohort. D Percentage distribution of 17q chromosome variation among the three subtypes in the GSE73517 cohort. E Percentage distribution of 11q chromosome variation among the three subtypes in the GSE73517 cohort. F Percentage distribution of 1p chromosome variation among the three subtypes in the GSE73517 cohort

Biological peculiarities analysis among the three subtypes

To elucidate the intrinsic biological features of the metabolism-related subtypes more comprehensively, a GSVA phenotypic analysis was performed. The mixed phenotype C1 subtype was characterized by upregulated mitotic spindle pathways and downregulated MYC targets (Fig. 5A, B). The most favorable prognosis C3 subtype exhibited upregulation in coagulation, KRAS, myogenesis, and UV-response pathways, along with downregulation of MYC targets, the G2M checkpoint, and E2F pathways (Fig. 5A, C,D). Notably, the worst prognostic C2 subtype displayed remarkably opposite characteristics compared to those of the C3 subtype, featuring upregulated MYC targets, G2M checkpoint, and E2F pathways as well as downregulated coagulation (Fig. 5A, C,D). Overall, these three subtypes exhibited transcriptional characteristics that were relatively consistent with the prognosis.

Fig. 5.

Fig. 5

Open in a new tab

Biological peculiarities analysis among the three subtypes. A GSVA phenotypic analysis the three subtypes presented by a heatmap. B PCA analysis of the mitotic-spindle signaling among the C1, C2, and C3 subtypes. C PCA analysis of the MYC target signaling among the C1, C2, and C3 subtypes. D PCA analysis of the coagulation signaling among the C1, C2, and C3 subtypes. B–D present PCA plots of pathway scores, where each point represents a single sample and the color intensity reflects the degree of activation of the indicated pathway in that sample, darker colors correspond to higher pathway activity

Immune landscape analysis of the three subtypes

Immunomodulators are of vital importance for cancer immunotherapy in clinical settings. Therefore, we further investigated the five major categories of 135 immunomodulators (21 antigen-presentation molecules, 39 chemokines molecules, 19 immuno-inhibitors molecules, 40 immuno-stimulators molecules, and 16 receptors molecules) in the three NB subtypes. Our analysis revealed significant differences in the transcriptional expression of the immunomodulators. Specifically, immune regulatory expression in C2 subtype patients was markedly suppressed, resulting in an immune-suppressed phenotype, followed by the C1 subtype, in which the antigen-presentation molecules were relatively active, but the immune-stimulators molecules were in an inhibited state, resulting in a mixed phenotype. In contrast, C3 subtype patients exhibit a higher expression pattern of immune regulators, presenting as an immune-activated phenotype (Fig. 6). In addition, we quantified the matrix and immune fractions in the immune microenvironments of subtypes C2 and C3 using the ESTAMITE software package. The results showed that immune and stromal Score of C3 subtype were higher than those of C2 subtype, and the expression of the immune check point inhibitor (ICI) gene was higher than that of C2 subtype. The infiltration degree of 22 immune cells in C3 subtypes (22/24, 91.67%) was higher than that in C2 subtype (P < 0.05) (Supplementary Fig. 1). Therefore, these results preliminarily suggest that the NB patients with the C3 subtype might potentially benefit from clinical immunotherapy, while those with the C2 subtype, characterized by significant immunosuppressive properties, may be prone to immune-related adverse events and therefore less suitable for this therapeutic approach.

Fig. 6.

Fig. 6

Open in a new tab

Distinct immune microenvironment infiltration among the three subtypes. Heatmap representation of quantified results of the immune microenvironment (21 antigen-presentation molecules, 39 chemokine molecules, 19 immuno-inhibitor molecules, 40 immuno-stimulator molecules, and 16 receptor molecules) across the three subtypes

Construction and validation of the NB prognostic model

To further explore the key genes that lead to the differences in prognosis, we calculated gene expressions in the different groups based on the TPM data and found that 390 genes were differentially expressed (P < 0.05) (Fig. 7A). Univariate Cox regression analysis was used to calculate the metabolic genes affecting the NB prognosis, and 36 metabolic differential genes with prognostic effects were further identified (P < 0.05). Kaplan-Meier survival analysis of the 36 genes in the GSE49710 cohort is shown in Supplementary Fig. 2, (all P < 0.001). Next, we used ten ML methods (Lasso, Enet, plsRcox, CoxBoost, StepCox, GBM, Ridge, RSF, survivor-SVM, and SuperPC) with 110 ML algorithms to construct 92 prognostic prediction models for the AE, GSE49710, and TARGET cohorts (Fig. 7B). Among these, the combination of the StepCox [forward] + RSF algorithm (C-index = 0.822) was the best modeling methods. The Prognostic Risk Score (RS) was calculated for patients with NB. This prognostic RS comprises 17 metabolism-related genes (POLD1, PRIM1, POLA2, RRM2, MTHFD2, NME4, POLR3D, UCK2, ACHE, INPP4B, ADCY1, AGPAT4, PLCD4, DPYD, PIK3CD, ENTPD3, and PDE11A) (Fig. 7C). The RSF importance score showed that UCK2 had the highest importance score (C2 subtype upregulated, whereas C3 subtype downregulated, importance = 15.9) (Fig. 7D). The revealed that UCK2 made a significantly contributed to the prognostic model. Kaplan-Meier analysis further showed that the models could effectively divide patients with NB into two prognostic RS groups (high and low) (log-rank test, P < 0.001) (Fig. 7E), and the ROC curves also showed that the prognostic RS had excellent specificity and sensitivity. The AUC for 1-, 3-, and 5-year survival rates were 0.852, 0.911, and 0.914, respectively (Fig. 7F). Moreover, two external NB cohorts (AE and TARGET) were used to further validate the reliability of the risk model (AE, P < 0.0001; TARGET, P = 0.00017) (Supplementary Fig. 3). In conclusion, we have established a reliable prognostic prediction model for NB based on UCK2 expression.

Fig. 7.

Fig. 7

Open in a new tab

Construction and validation of NB prognostic model. A Screening of significantly different metabolic-related genes between C2and C3 subtypes. Significant differentially expressed genes were determined with criteria set at: |log2FoldChange| >1, P < 0.05. B 92 prognostic prediction models were constructed by 10 multiple ML methods (Lasso, Enet, plsRcox, CoxBoost, StepCox, GBM, Ridge, RSF, Survivor-SVM, and SuperPC) with 110 ML algorithms in AE, GSE49710, and TARGET cohort. C Expression analysis of the 17 key modeling genes in the C2 and C3 subtypes. D The importance analysis of the 17 key modeling genes. E Kaplan-Meier analysis demonstrated that the models could effectively stratify GSE49710 cohort into two prognostic risk (high- and low-) groups. F Specificity and sensitivity of the prognostic models were analyzed at 1-, 3-, and 5-year survival using ROC curves

Validation of UCK2 in clinical tissues

Specifically, the histopathology of tissues revealed that two patients had well-differentiated, six had poorly differentiated, and two had moderately differentiated tumors. In terms of clinical stage, one patient was stage 1, two were stage 2, three were stage 3, four were stage 4, and none was stage 4 S. In addition, seven patients showed MYCN expression. We further investigated the expression of UCK2 in NB tissues using IHC staining, which indicated that UCK2 exhibited a higher expression in NB tissues than in ganglioneuroma tissues (Fig. 8). Notably, the positive expression of UCK2 was significantly higher than that of clinical histopathology, stages, and MYCN expression. The validation tissue data were in accordance with the above study and further confirmed that metabolism-related genes could be effectively used to distinguish NB progression.

Fig. 8.

Fig. 8

Open in a new tab

Validation of UCK2 in clinical tissues. Immunohistochemistry staining was used to evaluate the UCK2 expression in NB and ganglioneuroma tissues

Discussion

Several molecular subtypes of NB have been identified based on their gene expression levels using bioinformatics analyses [21]. The discovery and exploration of these molecular subtypes have provided insights into NB tumor heterogeneity. The molecular subtypes constructed based on transient expression levels of the transcriptome are often unstable and difficult to reproduce [15, 22] because gene expression in tissues is a dynamic process, and molecular subtypes based on static transcriptome profiles largely ignore dynamic changes in gene expression levels within biological systems. In contrast, biological networks remain relatively stable over time and across conditions and are thus more reliable for representing the biological state of tissues. Notably, network analysis is more robust and effective than single-gene methods, which use high-dimensional data [23, 24]. However, most network-based methods focus only on the gene nodes in biological networks and neglect the interactions between genes. We addressed this issue by introducing a gene sorting-based gene edge perturbation network algorithm that uses gene node information and includes critical information on the interactions within biological networks [15]. Genetic interactions are highly conserved in normal samples, but are extensively disrupted in diseased tissues. The relationships among genes in a network can be examined to characterize the changes in their interactions [25]. Therefore, the overall perturbation of all gene pairs in a background network can be used to effectively represent individual-level pathological conditions, define corresponding NB subtypes, and enhance our understanding of NB heterogeneity from a genetic interaction perspective.

We identified 948 metabolism-related genes in 876 NB tissue samples from three databases. A large-scale gene interaction perturbation network comprising 167,849 edges was constructed. We performed unsupervised cluster analysis of this network to classify patients with NB into three subtypes (C1-C3), each characterized by distinct clinical and molecular features.

Tumor cells typically reprogram their metabolism to meet the high-energy demands required for rapid proliferation, invasion, and metastasis. Therefore, tumor-associated metabolic reprogramming is a characteristic of tumors [26]. We identified metabolic reprogramming in NB. The C2 subtype was associated with the poorest prognosis; nucleotide metabolism (purine and pyrimidine pathways) was notably higher in this subtype than in subtypes C1 and C3. In contrast, the activation of nucleotide metabolism was lower in subtype C3. Increased nucleotide metabolism is a prerequisite for tumor initiation and development and is often closely related to the tumor cell type and genetic background. Severe metabolic alterations are typically accompanied by aggressive malignant behaviors such as uncontrolled proliferation, chemotherapy resistance, immune evasion, and metastasis [27]. Our findings are consistent with those of previous studies [28], which showed that higher levels of nucleotide metabolism often signify worse prognosis and more aggressive tumor phenotypes. Additionally, immune suppression was characterized by subtype C2, whereas immune activation was characterized by subtype C3. This finding further confirms that metabolic reprogramming in NB plays a key role in maintaining tumorigenesis and metastasis and more broadly affects the antitumor immune response by releasing metabolites and influencing the expression levels of immune molecules [29]. Targeting key enzymes in tumor cell nucleotide biosynthesis is a novel and promising therapeutic strategy because of the role of nucleotide metabolism in regulating TIM and immune function [26]. Extracellular nucleotide ATP stimulates antitumor immunity by activating purine receptors, whereas adenosine is an effective immunosuppressant. Targeting NDUFB10, a key mitochondria-related gene, is a potential therapeutic strategy for patients with LUAD [30]. Nucleotide metabolism may be a target of TIMs in cancer treatment [31]. Therefore, tumor treatments targeting nucleotides may become a key approach for improving the prognosis of patients with the C2 subtype of NB, despite the low efficacy of traditional immunotherapy.

The reprogramming of amino acid and glucose metabolism is another key aspect of tumor metabolic reprogramming. Significant reprogramming of amino acid and glucose metabolism has been observed in patients with C2 subtype, specifically manifesting as enhanced levels of cysteine, methionine, and glucose metabolism products (glyoxylate and dicarboxylate). High MYCN levels meet the large demand of tumor cells for cysteine through uptake and transculturation pathways, mediating the antioxidant response of tumor cells and maintaining the redox balance, which are crucial for reducing extensive lipid peroxidation and ferroptosis in NB cells [32]. The C2 subtype has several adverse clinical features, such as high risk, MYCN amplification, increased 17q gain, and 1p deletion. In Addition, tumor cells inhibit T-cell immunity by competitively consuming amino acids in the microenvironment. The large quantities of amino acids obtained interfere with T cells by regulating glucose and lipid metabolism, producing a tumor-immunosuppressive state, and promoting tumor occurrence and survival [33]. Therefore, antitumor therapeutic strategies that target amino acid metabolism are attracting increasing attention. For example, amino acid antagonists target amino acid transporters, key enzymes in amino acid metabolism, and common downstream pathways of amino acid metabolism, in addition to playing crucial roles in maintaining tumor growth. Advances in technology and research have revealed the broad importance of amino acid metabolic reprogramming in the regulation of antitumor immune responses [33, 34]. Therefore, antitumor therapies that combine metabolism and immunotherapy should be developed for patients with C2 subtype NB.

The rapid development of artificial intelligence has enabled the development of tumor prognosis prediction models [35] and enhancing diagnostic accuracy [36, 37] based on large-scale datasets using multiple ML and deep learning algorithms. Li et al. constructed six ML algorithms based on univariate and multivariate logistic analyses for population-based and external validation and developed 101 algorithms to construct precise and personalized predictive prognostic models for NB (AUC < 0.85) [38]. In the present study, we screened NB markers by constructing a perturbation network algorithm for metabolism-related genes. Next, we used ten ML methods to construct 110 algorithms to produce a model that effectively predicts the prognosis of NB. The predictive AUCs were 0.852, 0.911, and 0.914 for 1-, 3-, and 5-year survival rates, respectively. UCK2 was the primary biomarker in this model. UCK2 is a rate-limiting enzyme in the pyrimidine salvage synthesis pathway, which is overexpressed in various solid and hematological cancers [39] and is closely related to poor prognosis by promoting cell proliferation and migration in lung [40] and liver [41] cancers. The catalytic activity of UCK2 has been exploited to develop several cytotoxic ribonucleoside analogues, such as TAS-106 and RX-3117, for UCK2-mediated cancer chemotherapy [42]. Using IHC, we further verified that UCK2 expression level is high in patients with NB and gliomas. Although our model demonstrated robustness across three independent transcriptomic datasets, the current lack of open-access neuroblastoma cohorts with associated survival data restricts further prospective validation. Additional clinical samples will be incorporated as such resources become available. These findings initial indicate that UCK2 is a therapeutic target for NB.

We constructed a model specifically for predicting the prognosis of patients with NB using new analytical methods and ML; however, this model has some limitations that should be addressed. First, we were unable to fully validate the model in clinical practice because of the limited sample size. We encourage the participation of more research centers to optimize the model to benefit more patients. Second, the prospects of using UCK2 in treating NB were not explored, and the other 16 genes should be also analyzed and will be our focus in future studies. Finally, patients with C2 subtype NB have a poor prognosis. Methods for the incorporation of metabolic reprogramming into the clinical treatment of NB have not yet been developed. These aspects are key to future research, and the findings may affect the prognosis and survival of patients with NB.

In conclusion, metabolism-related gene perturbation networks and unsupervised clustering analyses were used to stratify NB patients into three distinct subtypes: C1, with an intermediate prognosis; C2, with an unfavorable prognosis; and C3, with a favorable prognosis. A high-resolution classification model (StepCox [forward] + RSF algorithm) was established using multiple ML algorithms to predict the prognoses of patients with NB. This method provides precise guidance for clinical decision making. The metabolic-related NB subtypes were related to the current clinical prognostic criteria for NB, and the genes used for classification minimally overlapped with the characteristic genes identified in previous studies. This study revealed substantial molecular convergence and distinct biological heterogeneity among patients with NB. This insight indicates that high-resolution classification increases the efficacy of managing patients with NB as well as advances the research on and application of critical biomarkers in NB therapy.

Supplementary Information

12887_2026_6512_MOESM1_ESM.xlsx (145.3KB, xlsx)

Supplementary Material 1: Supplementary Table 1. Total of 948 genes and 41 metabolism-related pathways. S1 Fig. Comparison of the matrix fraction and immune fraction in the immune microenvironment (ImmuneScore, StromalScore, ICI, TIME) between the C2 and C3 subtypes. S2 Fig. Kaplan-Meier survival analysis of 36 genes in the GSE49710 cohort. S3 Fig. Validation of the prediction model in two external NB cohorts (AE and TARGET).

12887_2026_6512_MOESM2_ESM.tif (14.6MB, tif)

Supplementary Material 2: Supplementary Figure 1. Comparison of the matrix fraction and immune fraction in the immune microenvironment (ImmuneScore, StromalScore, ICI, TIME) between the C2 and C3 subtypes.

12887_2026_6512_MOESM3_ESM.tif (13.2MB, tif)

Supplementary Material 3: Supplementary Figure 2. Kaplan-Meier survival analysis of 36 genes in the GSE49710 cohort.

12887_2026_6512_MOESM4_ESM.tif (9.2MB, tif)

Supplementary Material 4: Supplementary Figure 3. Validation of the prediction model in two external NB cohorts. A Kaplan-Meier survival curve showed the survival status of High- and Low- risk subtypes in AE cohort. B Kaplan-Meier survival curve showed the survival status of High- and Low- risk subtypes TARGET cohort.

Acknowledgements

We thank the KEGG database for providing pathway information and all the participants who took part in our study, and medical record managers for their supports. We would like to thank Editage (www.editage.cn) for English language editing.

Abbreviations

NB

Neuroblastoma

HVA

Homovanillic acid

VMA

Vanillylmandelic acid

ML

Machine learning

TIME

Tumor immune microenvironment

GEO

Gene Expression Omnibus

SNP

Single nucleotide polymorphism

PCA

Principal Component Analysis

GO

Gene Ontology

KEGG

Kyoto Encyclopedia of Genes and Genomes

ssGSEA

Single-sample Gene Set Enrichment Analysis

K-M

Kaplan-Meier

ROC

Receiver Operating Characteristic

AUC

Area Under the Curve

UMAP

Unified Manifold Approximation and Projection

Authors’ contributions

CY and JW designed and supervised the study. XL collected the data and wrote the drafting manuscript. XH did the bioinformatics analysis. QC did the experiment and literature search. CY and JW revised the manuscript. All authors contributed to the manuscript.

Funding

This study was supported by the Natural Science Foundation of Wuhan Municipal Health Commission (WZ22Q08, WZ24B44). Hubei Provincial Natural Science Foundation of China (2024AFB968) and Soaring Plan of Youth Talent Development in Wuhan Children’s Hospital.

Data availability

The datasets used and/or analyzed during the current study are available on reasonable request from corresponding author.

Declarations

Ethics approval and consent to participate

This is a low-risk experimental verification study. The authors declared that all the procedures followed were in accordance with the ethical standards of the responsible committee on human experimentation (national and institutional). The experimental protocol was reviewed and approved by the Ethics Committee of Wuhan Children’s Hospital (2021R134-E01). The requirement for individual informed consent was waived by the ethics committee listed above because this study used currently existing samples during routine medical care and did not pose any additional risks to the patients. All patient data were anonymized prior to the analysis. Any administrative permissions and/or licenses were acquired by the ethics committee of the Wuhan Children’s Hospital, Huazhong University of Science & Technology to access the data used in our research. All methods were performed in accordance with the relevant guidelines and regulations.

Consent for publication

All authors approved the manuscript and gave their consent for submission and publication.

Competing interests

The authors declare no competing interests.

Footnotes

Publisher’s note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Xin Liu, Xin Hu and Qinzhen Cai contributed equally to this work.

Contributor Information

Chunhui Yuan, Email: chunhuii.yuen@whu.edu.cn.

Jun Wang, Email: sywj0928@126.com.

References

  • 1.Zafar A, Wang W, Liu G, Wang X, Xian W, McKeon F, et al. Molecular targeting therapies for neuroblastoma: progress and challenges. Med Res Rev. 2021;41(2):961–1021. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 2.Qiu B, Matthay KK. Advancing therapy for neuroblastoma. Nat Rev Clin Oncol. 2022;19(8):515–33. [DOI] [PubMed] [Google Scholar]
  • 3.Matthay KK, Maris JM, Schleiermacher G, Nakagawara A, Mackall CL, Diller L, et al. Neuroblastoma. Nat Rev Dis Primers. 2016;2:16078. [DOI] [PubMed] [Google Scholar]
  • 4.Berko ER, Naranjo A, Daniels AA, McNulty SN, Krytska K, Druley T, et al. Frequency and clinical significance of clonal and subclonal driver mutations in High-Risk neuroblastoma at diagnosis: A children’s oncology group study. J Clin Oncology: Official J Am Soc Clin Oncol. 2025;43(14):1673–84. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 5.Owens C, Irwin M. Neuroblastoma: the impact of biology and cooperation leading to personalized treatments. Crit Rev Clin Lab Sci. 2012;49(3):85–115. [DOI] [PubMed] [Google Scholar]
  • 6.Matser YAH, Verly IRN, van der Ham M, de Sain-van der Velden MGM, Verhoeven-Duif NM, Ash S, et al. Optimising urinary catecholamine metabolite diagnostics for neuroblastoma. Pediatr Blood Cancer. 2023;70(6):e30289. [DOI] [PubMed] [Google Scholar]
  • 7.Irwin MS, Naranjo A, Zhang FF, Cohn SL, London WB, Gastier-Foster JM, et al. Revised neuroblastoma risk classification system: a report from the Children’s Oncology Group. J Clin Oncol. 2021;39(29):3229–41. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 8.Ackermann S, Cartolano M, Hero B, Welte A, Kahlert Y, Roderwieser A, et al. A mechanistic classification of clinical phenotypes in neuroblastoma. Science. 2018;362(6419):1165–70. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 9.Alburquerque-Bejar JJ, Navajas-Chocarro P, Saigi M, Ferrero-Andres A, Morillas JM, Vilarrubi A, et al. MYC activation impairs cell-intrinsic IFNgamma signaling and confers resistance to anti-PD1/PD-L1 therapy in lung cancer. Cell Rep Med. 2023;4(4):101006. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 10.Barone G, Anderson J, Pearson AD, Petrie K, Chesler L. New strategies in neuroblastoma: therapeutic targeting of MYCN and ALK. Clin Cancer Res. 2013;19(21):5814–21. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 11.Berlak M, Tucker E, Dorel M, Winkler A, McGearey A, Rodriguez-Fos E, et al. Mutations in ALK signaling pathways conferring resistance to ALK inhibitor treatment lead to collateral vulnerabilities in neuroblastoma cells. Mol Cancer. 2022;21(1):126. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 12.Sakowicz-Burkiewicz M, Pawelczyk T, Zysk M. Role of energy metabolism in the progression of neuroblastoma. Int J Mol Sci. 2021;22(21):11421. 10.3390/ijms222111421. [DOI] [PMC free article] [PubMed]
  • 13.Colletti M, Petretto A, Galardi A, Di Paolo V, Tomao L, Lavarello C, et al. Proteomic analysis of Neuroblastoma-Derived exosomes: new insights into a metastatic signature. Proteomics. 2017;17:23–4. [DOI] [PubMed] [Google Scholar]
  • 14.Yoshida GJ. Beyond the warburg effect: N-Myc contributes to metabolic reprogramming in cancer cells. Front Oncol. 2020;10:791. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 15.Liu Z, Weng S, Dang Q, Xu H, Ren Y, Guo C, et al. Gene interaction perturbation network Deciphers a high-resolution taxonomy in colorectal cancer. eLife. 2022;11:e81114. 10.7554/eLife.81114. [DOI] [PMC free article] [PubMed]
  • 16.Seiler M, Huang CC, Szalma S, Bhanot G. ConsensusCluster: a software tool for unsupervised cluster discovery in numerical data. Omics. 2010;14(1):109–13. [DOI] [PubMed] [Google Scholar]
  • 17.Kanehisa M, Goto S. KEGG: kyoto encyclopedia of genes and genomes. Nucleic Acids Res. 2000;28(1):27–30. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 18.Fortunel NO, Otu HH, Ng HH, Chen J, Mu X, Chevassut T, et al. Comment on ‘Stemness’: transcriptional profiling of embryonic and adult stem cells and a stem cell molecular signature. Science. 2003;302(5644):393. author reply. [DOI] [PubMed] [Google Scholar]
  • 19.Lu X, Meng J, Su L, Jiang L, Wang H, Zhu J, et al. Multi-omics consensus ensemble refines the classification of muscle-invasive bladder cancer with stratified prognosis, tumour microenvironment and distinct sensitivity to frontline therapies. Clin Transl Med. 2021;11(12):e601. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 20.Malta TM, Sokolov A, Gentles AJ, Burzykowski T, Poisson L, Weinstein JN, et al. Machine learning identifies stemness features associated with oncogenic dedifferentiation. Cell. 2018;173(2):338–54. e15. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 21.Korber V, Stainczyk SA, Kurilov R, Henrich KO, Hero B, Brors B, et al. Neuroblastoma arises in early fetal development and its evolutionary duration predicts outcome. Nat Genet. 2023;55(4):619–30. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 22.Fischer DS, Fiedler AK, Kernfeld EM, Genga RMJ, Bastidas-Ponce A, Bakhti M, et al. Inferring population dynamics from single-cell RNA-sequencing time series data. Nat Biotechnol. 2019;37(4):461–8. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 23.Luo R, Yan J, Oh JW, Xi W, Shigaki D, Wong W, et al. Dynamic network-guided CRISPRi screen identifies CTCF-loop-constrained nonlinear enhancer gene regulatory activity during cell state transitions. Nat Genet. 2023;55(8):1336–46. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 24.Zhang H, Zhang G, Xu P, Yu F, Li L, Huang R, et al. Optimized dynamic network biomarker deciphers a high-resolution heterogeneity within thyroid cancer molecular subtypes. Med Res. 2025;1(1):10–31. [Google Scholar]
  • 25.Yi S, Lin S, Li Y, Zhao W, Mills GB, Sahni N. Functional variomics and network perturbation: connecting genotype to phenotype in cancer. Nat Rev Genet. 2017;18(7):395–410. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 26.Shuvalov O, Petukhov A, Daks A, Fedorova O, Vasileva E, Barlev NA. One-carbon metabolism and nucleotide biosynthesis as attractive targets for anticancer therapy. Oncotarget. 2017;8(14):23955–77. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 27.Mullen NJ, Singh PK. Nucleotide metabolism: a pan-cancer metabolic dependency. Nat Rev Cancer. 2023;23(5):275–94. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 28.Luo D, Wang H, Zeng Z, Chen J, Wang H. Integrated bioinformatics analysis of nucleotide metabolism based molecular subtyping and biomarkers in lung adenocarcinoma. Front Immunol. 2024;15:1430171. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 29.Xia L, Oyang L, Lin J, Tan S, Han Y, Wu N, et al. The cancer metabolic reprogramming and immune response. Mol Cancer. 2021;20(1):28. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 30.Zhang P, Zhang M, Liu J, Zhou Z, Zhang L, Luo P, et al. Mitochondrial pathway signature (MitoPS) predicts immunotherapy response and reveals NDUFB10 as a key immune regulator in lung adenocarcinoma. J Immunother Cancer. 2025. 10.1136/jitc-2025-012069. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 31.Madsen HB, Peeters MJ, Straten PT, Desler C. Nucleotide metabolism in the regulation of tumor microenvironment and immune cell function. Curr Opin Biotechnol. 2023;84:103008. [DOI] [PubMed] [Google Scholar]
  • 32.Alborzinia H, Florez AF, Kreth S, Bruckner LM, Yildiz U, Gartlgruber M, et al. MYCN mediates cysteine addiction and sensitizes neuroblastoma to ferroptosis. Nat Cancer. 2022;3(4):471–85. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 33.Zheng Y, Yao Y, Ge T, Ge S, Jia R, Song X, et al. Amino acid metabolism reprogramming: shedding new light on T cell anti-tumor immunity. J Exp Clin Cancer Res. 2023;42(1):291. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 34.Yang L, Chu Z, Liu M, Zou Q, Li J, Liu Q, et al. Amino acid metabolism in immune cells: essential regulators of the effector functions, and promising opportunities to enhance cancer immunotherapy. J Hematol Oncol. 2023;16(1):59. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 35.Tran KA, Kondrashova O, Bradley A, Williams ED, Pearson JV, Waddell N. Deep learning in cancer diagnosis, prognosis and treatment selection. Genome Med. 2021;13(1):152. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 36.Yifan C, Jianfeng S, Jun P. Development and validation of a random forest diagnostic model of acute myocardial infarction based on ferroptosis-related genes in circulating endothelial cells. Front Cardiovasc Med. 2021;8:663509. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 37.Sherpa D, Bhowmick C, Pavan T, Rajwade DA, Halder S, Mitra I, et al. Classification of idiopathic recurrent spontaneous miscarriage using FTIR and raman spectroscopic fusion technology. Syst Biol Reprod Med. 2024;70(1):228–39. [DOI] [PubMed] [Google Scholar]
  • 38.Li S, Wang J, Zhang Z, Ren C, He D. Individual risk and prognostic value prediction by interpretable machine learning for distant metastasis in neuroblastoma: a population-based study and an external validation. Int J Med Inform. 2025;196:105813. [DOI] [PubMed] [Google Scholar]
  • 39.Watanabe T, Yamamoto Y, Kurahashi Y, Kawasoe K, Kidoguchi K, Ureshino H, et al. Reprogramming of pyrimidine nucleotide metabolism supports vigorous cell proliferation of normal and malignant T cells. Blood Adv. 2024;8(6):1345–58. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 40.Wu Y, Jamal M, Xie T, Sun J, Song T, Yin Q, et al. Uridine-cytidine kinase 2 (UCK2): a potential diagnostic and prognostic biomarker for lung cancer. Cancer Sci. 2019;110(9):2734–47. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 41.Sha N, Zhou B, Hou G, Xi Z, Wang W, Yan M, et al. The protection of UCK2 protein stability by GART maintains pyrimidine salvage synthesis for HCC growth under glucose limitation. Oncogene. 2025;44(16):1078–92. [DOI] [PubMed] [Google Scholar]
  • 42.Fu Y, Wei XD, Guo L, Wu K, Le J, Ma Y, et al. The metabolic and non-metabolic roles of UCK2 in tumor progression. Front Oncol. 2022;12:904887. [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

12887_2026_6512_MOESM1_ESM.xlsx (145.3KB, xlsx)

Supplementary Material 1: Supplementary Table 1. Total of 948 genes and 41 metabolism-related pathways. S1 Fig. Comparison of the matrix fraction and immune fraction in the immune microenvironment (ImmuneScore, StromalScore, ICI, TIME) between the C2 and C3 subtypes. S2 Fig. Kaplan-Meier survival analysis of 36 genes in the GSE49710 cohort. S3 Fig. Validation of the prediction model in two external NB cohorts (AE and TARGET).

12887_2026_6512_MOESM2_ESM.tif (14.6MB, tif)

Supplementary Material 2: Supplementary Figure 1. Comparison of the matrix fraction and immune fraction in the immune microenvironment (ImmuneScore, StromalScore, ICI, TIME) between the C2 and C3 subtypes.

12887_2026_6512_MOESM3_ESM.tif (13.2MB, tif)

Supplementary Material 3: Supplementary Figure 2. Kaplan-Meier survival analysis of 36 genes in the GSE49710 cohort.

12887_2026_6512_MOESM4_ESM.tif (9.2MB, tif)

Supplementary Material 4: Supplementary Figure 3. Validation of the prediction model in two external NB cohorts. A Kaplan-Meier survival curve showed the survival status of High- and Low- risk subtypes in AE cohort. B Kaplan-Meier survival curve showed the survival status of High- and Low- risk subtypes TARGET cohort.

Data Availability Statement

The datasets used and/or analyzed during the current study are available on reasonable request from corresponding author.

Articles from BMC Pediatrics are provided here courtesy of BMC

RESOURCES